With the advancements of various autonomous car projects aiming to achieve SAE Level 5, real-time detection of traffic signs in real-life scenarios has become a highly relevant problem for the industry. Even though a great progress has been achieved in this field, there is still no clear consensus on what the state-of-the-art in this field is. Moreover, it is important to develop and test systems in various regions and conditions. This is why the "Ice Vision" competition has focused on the detection of Russian traffic signs in winter conditions. The IceVisionSet dataset used for this competition features real-world collection of lossless frame sequences with traffic sign annotations. The sequences were collected in varying conditions, including: different weather, camera exposure, illumination and moving speeds. In this work we describe the competition and present the solutions of the 3 top teams.